bf16: true
cutoff_len: 2048
dataset: prithivMLmods/Deepthink-Reasoning-Ins,open-thoughts/OpenThoughts-114k
dataset_dir: data
ddp_timeout: 180000000
do_train: true
eval_steps: 200
eval_strategy: steps
finetuning_type: full
flash_attn: auto
gradient_accumulation_steps: 8
include_num_input_tokens_seen: true
learning_rate: 0.0001
logging_steps: 1
lr_scheduler_type: polynomial
max_grad_norm: 1.0
max_samples: 57000000
model_name_or_path: Qwen/Qwen2.5-0.5B-Instruct
num_train_epochs: 1.0
optim: sgd
output_dir: saves\Qwen2.5-0.5B-Instruct\full\20-02-2025
packing: false
per_device_eval_batch_size: 1
per_device_train_batch_size: 1
plot_loss: true
preprocessing_num_workers: 16
report_to: none
save_steps: 200
stage: sft
template: qwen
trust_remote_code: true
val_size: 1
warmup_steps: 0